PHUMA: Physically-Grounded Humanoid Locomotion Dataset

2025-10-30

AI Summary

Existing methods face a trade-off between scalability and physical plausibility: high-fidelity motion-capture datasets (e.g., AMASS) are limited in scale, while large-scale web-video datasets (e.g., Humanoid-X) suffer from pervasive physical artifacts—floating, interpenetration, and foot slipping. This work introduces the first physics-constrained framework for constructing large-scale, physically plausible humanoid robot motion datasets. Leveraging high-accuracy pose estimation, it extracts motions from internet videos and enforces physical consistency via joint-limit constraints, ground-contact modeling, and slip-resistant motion retargeting. Crucially, it achieves unprecedented scalability *without* compromising physical realism. Experiments demonstrate substantial improvements over Humanoid-X and AMASS on unseen-motion imitation and pelvis trajectory tracking, yielding enhanced policy stability and cross-motion generalization.

Technology Category

• Computer Vision: Motion & Tracking
• Intelligent Robots: Motion and Path Planning
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines

Abstract

Motion imitation is a promising approach for humanoid locomotion, enabling agents to acquire humanlike behaviors. Existing methods typically rely on high-quality motion capture datasets such as AMASS, but these are scarce and expensive, limiting scalability and diversity. Recent studies attempt to scale data collection by converting large-scale internet videos, exemplified by Humanoid-X. However, they often introduce physical artifacts such as floating, penetration, and foot skating, which hinder stable imitation. In response, we introduce PHUMA, a Physically-grounded HUMAnoid locomotion dataset that leverages human video at scale, while addressing physical artifacts through careful data curation and physics-constrained retargeting. PHUMA enforces joint limits, ensures ground contact, and eliminates foot skating, producing motions that are both large-scale and physically reliable. We evaluated PHUMA in two sets of conditions: (i) imitation of unseen motion from self-recorded test videos and (ii) path following with pelvis-only guidance. In both cases, PHUMA-trained policies outperform Humanoid-X and AMASS, achieving significant gains in imitating diverse motions. The code is available at https://davian-robotics.github.io/PHUMA.

Problem

• Addresses physical artifacts in motion imitation from video data
• Eliminates floating, penetration, and foot skating in humanoid locomotion
• Creates physically reliable motions while maintaining large-scale diversity

Innovation

• Leverages large-scale human videos for data generation
• Addresses physical artifacts via physics-constrained retargeting
• Ensures ground contact and eliminates foot skating